Extreme Fire Severity Patterns in Topographic ... - ScienceOpen

0 downloads 0 Views 700KB Size Report
Jan 22, 2014 - Mediterranean Pine Forests. Judit Lecina-Diaz1*, Albert Alvarez1, Javier Retana1,2 ...... Propuesta metodológica. In: Juaritsi J, Moro I, editors.
Extreme Fire Severity Patterns in Topographic, Convective and Wind-Driven Historical Wildfires of Mediterranean Pine Forests Judit Lecina-Diaz1*, Albert Alvarez1, Javier Retana1,2 1 CREAF, Cerdanyola del Valle`s, Catalonia, Spain, 2 Departament de Biologia Animal, Biologia Vegetal i Ecologia, Universitat Auto`noma de Barcelona, Cerdanyola del Valle`s, Catalonia, Spain

Abstract Crown fires associated with extreme fire severity are extremely difficult to control. We have assessed fire severity using differenced Normalized Burn Ratio (dNBR) from Landsat imagery in 15 historical wildfires of Pinus halepensis Mill. We have considered a wide range of innovative topographic, fuel and fire behavior variables with the purposes of (1) determining the variables that influence fire severity patterns among fires (considering the 15 wildfires together) and (2) ascertaining whether different variables affect extreme fire severity within the three fire types (topographic, convective and wind-driven fires). The among-fires analysis showed that fires in less arid climates and with steeper slopes had more extreme severity. In less arid conditions there was more crown fuel accumulation and closer forest structures, promoting high vertical and horizontal fuel continuity and extreme fire severity. The analyses carried out for each fire separately (within fires) showed more extreme fire severity in areas in northern aspects, with steeper slopes, with high crown biomass and in climates with more water availability. In northern aspects solar radiation was lower and fuels had less water limitation to growth which, combined with steeper slopes, produced more extreme severity. In topographic fires there was more extreme severity in northern aspects with steeper slopes and in areas with more water availability and high crown biomass; in convectiondominated fires there was also more extreme fire severity in northern aspects with high biomass; while in wind-driven fires there was only a slight interaction between biomass and water availability. This latter pattern could be related to the fact that wind-driven fires spread with high wind speed, which could have minimized the effect of other variables. In the future, and as a consequence of climate change, new zones with high crown biomass accumulated in non-common drought areas will be available to burn as extreme severity wildfires. Citation: Lecina-Diaz J, Alvarez A, Retana J (2014) Extreme Fire Severity Patterns in Topographic, Convective and Wind-Driven Historical Wildfires of Mediterranean Pine Forests. PLoS ONE 9(1): e85127. doi:10.1371/journal.pone.0085127 Editor: Han Y. H. Chen, Lakehead University, Canada Received June 30, 2013; Accepted November 22, 2013; Published January 22, 2014 Copyright: ß 2014 Lecina-Diaz et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Funding: This study has been partially funded by the project MONTES CONSOLIDER CSD-2008-00040. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing Interests: The authors have declared that no competing interests exist. * E-mail: [email protected]

derived from remote sensing, which capture the substantial spectral changes that fire causes by consuming vegetation, destroying leaf chlorophyll, exposing soil and charring stems [9]. One of the most common variables measured by remote sensing is fire severity, defined as the degree of fire-induced environmental change on vegetation immediately after fire [10]. High values of fire severity are commonly related to crown fires and are important to validate fire risk maps, fire behavior models and management effectiveness [11,12]. Moreover, the use of remotesensing data allows the quantification of fire patterns over time and space, in particular the study of historical wildfires without available field data [13]. Landsat image data has been shown to classify accurately a large variety of landscapes, including the heterogeneous Mediterranean landscapes [14,15] and this imagery is usually transformed into indices (such as Normalized Difference Vegetation Index (NDVI) [16] or the Normalized Burn Ratio (NBR) [17]) by rationing spectral bands to assess fire severity [18]. Fire severity depends on the combination of physical variables, weather and fuels. Concerning physical variables, in some studies topography has been shown to affect the pattern of fire severities [19,20], whereas in other studies this effect is not clear and

Introduction Forest fires are common in many parts of the world, including the Mediterranean ecosystems [1]. Depending on the strata burnt, wildfires could be classified into ground, surface and crown fires [2] . Crown fires are those that burn in elevated canopy fuels, which mainly include active crown fires, when fuel and weather allow fire to spread continuously between tree crowns involving the entire surface-canopy complex, and passive crown fires, when one tree or a group of trees burns individually without a solid flame consistently maintained in the canopy [2,3]. Crown fires are extremely difficult to control due to their high rates of spread, intensity, flame lengths, spotting and fire severity, and are the major concern for fire managers and firefighters on safety, fire suppression and environmental developments [2,4]. Given the inherent risks associated with crown fires, some experimental studies that obtained direct information from fires have been focused on low-intensity fires [4,5], whereas measurements of extreme fire behavior associated with crown fires are more limited [6–8]. A widely used alternative method to obtain information from wildfires is based on multitemporal indices PLOS ONE | www.plosone.org

1

January 2014 | Volume 9 | Issue 1 | e85127

Factors Determining Extreme Fire Severity Patterns

coincident [21,22]. Weather variation is strongly related to fire severity. Specifically, low relative humidity, strong surface wind, unstable air and drought are described as the four critical weather elements of extreme fire behavior [6]. Regarding fuels, fuel moisture and forest structure are essential in determining the extreme fire behavior associated with crown fires [23]. But the way that fuels are related to fire severity is not obvious, as some authors suggest a clear relationship between both, even under drought conditions [13,24], whereas in other studies, especially under infrequent high and mixed-severity fire regimes, the role of fuels is reduced [25,26]. The European Project ‘‘Fire Paradox’’[27] analyzed the spread of fire in historical wildfires and showed that there were similar spread schemes dominated by common factors (e.g. wind direction and speed). Depending on the spread scheme and the dominant spread factor, three fire types were defined: convection or plumedominated fires, wind-driven fires and topographic fires [28,29]. Firstly, convection or plume-dominated fires are characterized by the accumulation of high quantity of available fuels and atmospheric instability. This fire type has such a high intensity and extreme behavior that produces its own fire environment and generates massive spotting. Secondly, wind-driven fires follow the speed and direction of strong winds when the meteorological window that produces the fire conditions is maintained, with the same intensity and velocity during day and night. In both of them, small changes in the landscape have little influence in the direction and behavior of these fire types, especially under extreme meteorological conditions. In contrast, topographic fires are dominated by local winds caused by slope and differences in solar heating of the earth surface (i.e. sea breeze, land breeze, valley and slope winds). The direction of this fire type changes with topography (e.g. hydrographic basins, main valley), and it has high intensity during the day and low intensity at night [28,29]. In the latter fire type, wildfire is more sensitive to small changes, thus little variations of topographical wind, slope or aspect have higher influence on fire behavior. The combination of two or three fire types in the same wildfire might be common in North America, Canada and Australia, since fire usually burns during many days or months and involves large areas of the landscape. Nevertheless, the majority of wildfires in Europe burn for 48 hours or less, thus fire has fewer opportunities to flip from one fire type to another. Moreover, the characterization of these three fire types allows the identification of the operational opportunities for the suppression systems [28,29]. Finally, these fire types are linked to meteorological conditions and topographical areas where they usually burn; thus, it is possible to know the risk of having one or another in the landscape depending on the meteorological forecast. The number of severe wildfires and their recurrence have increased during recent years in the Mediterranean Basin, leading to an increase in wildfires characterized by crown fire and extreme fire behavior [30,31]. Aleppo pine (Pinus halepensis Mill.) forests have been those most affected by high severity crown fires [31]. Pinus halepensis is one the most abundant conifers in the Mediterranean Basin, mostly at low elevations [32]. It is a serotinuos and not self-pruning species, with high vertical and horizontal continuity, which constitutes a highly flammable material that amplifies fire intensity from low-intensity surface fires to high-intensity crown fires [32,33]. The present study is based on 15 historical wildfires of P. halepensis that occurred in Catalonia (NE Spain) in the period 2000–2007. The studied fires corresponded to the three fire types described above (i.e., topographic, convective and wind-driven fires) and included a wide range of fuel, weather and physical variables. The objectives PLOS ONE | www.plosone.org

of the study are: (1) to determine the variables that influence extreme fire severity among fires (considering the 15 wildfires together), and (2) to ascertain which variables affect extreme fire severity within each fire (15 wildfires of the three different fire types). To achieve these aims we have integrated the information available from firefighter reports and from different data bases generating new variables, some of which had never previously been considered.

Materials and Methods Study area The 15 fires studied were located in Catalonia, Northeastern Spain, between 40u579 and 42u179 latitude North and 0u229 and 3u049 longitude East (Figure 1). The climate is Mediterranean, characterized by mild winters and hot and dry summers, although Catalonia encompasses a large climatic gradient [34,35]. Mean annual temperature ranged from 11.1 to 16.3uC, mean temperature of the hottest season (summer) from 19.3 to 24.1uC and mean temperature of the coldest season (winter) from 4.2 to 9.6uC. Mean annual precipitation ranged from 341 to 819 mm and mean precipitation of the driest season (summer) from 61 to 200 mm (Digital Climatic Atlas of Catalonia; [36,37]). The analyzed fires occurred between 2000 and 2007 and covered the three fire types according to the fire spread pattern: topographic, convection-dominated and wind-driven wildfires (Table S1). The selection of fires was determined by four conditions: (i) availability of cloud-free Landsat imagery before and after the fire, (ii) minimum forest cover within the fire of 30%, (iii) homogeneity in the fire type; and (iv) availability of information about fire behavior and fire effects from firefighter reports.

Computing and mapping fire severity Imagery used was obtained from Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+), which was geometrically corrected following the Pala` and Pons method [38], resulting in 20 m pixel size. Afterwards, a radiometric correction was applied to convert digital numbers (DN) to reflectance values, using a digital elevation model and parameters of exoatmospheric solar irradiance, atmospheric optical depth and sensor calibration [39]. The time between pre- and post-fire imagery was as short as possible (or near-anniversary date, [40]) in order to avoid phenologic changes of vegetation (Table S1). We used the differenced Normalized Burn Ratio (dNBR) [41,42], calculated from the Normalized Burn Ratio (NBR) for each image using bands 4 and 7, as shown in equation 1. The delta NBR (dNBR) was computed using the pre-fire minus the post-fire value of NBR (NBRpre and NBRpost in equation 2). The GISsoftware used was MiraMon 7.0 [43]. NBR~1000 

ðBand4{Band7) (Band4zBand7)

dNBR~NBRpre {NBRpost

ð1Þ

ð2Þ

In our study, we used field data from plots sampled in the Ventallo´ wildfire (fire number 14 on Figure 1, [12]) to obtain the severity thresholds. In this wildfire, we considered three severity levels within individual trees: (1) green trees, with at least 20% green crown; (2) scorched trees, which had less than 20% green 2

January 2014 | Volume 9 | Issue 1 | e85127

Factors Determining Extreme Fire Severity Patterns

Figure 1. Names and location of the fires studied. Black points indicate the center of the fire and inner lines indicate regions. doi:10.1371/journal.pone.0085127.g001

(ii) in the second level, within fires, the variable was fire severity analyzed for each wildfire separately, which was categorical with two levels: not extreme severity (including both scorched and green severity) and extreme severity (charred severity).

crown (although most of them were completely scorched); and (3) charred trees, which were skeletons mainly consumed without fine materials on the tree or on the ground [12]. Afterwards, three categories of severity at the plot level (20 m diameter) were defined: (1) green plots/low severity (at least 50% of green trees); (2) scorched plots/moderate severity (at least 50% of scorched trees and not more than 25% of green trees); and (3) charred plots/extreme severity (almost 100% of charred trees) [12]. We calculated the average dNBR value of a 363 pixel window centered in the plots of this fire and we obtained a data base containing the dNBR values and their corresponding severity level, thus defining ranges of dNBR and their number of charred and green-scorched plots (we grouped low and moderate severity). Afterwards, we defined the possible severity thresholds and we calculated the percentage of correct classification in every threshold (the percentage of correctly defined charred or greenscorched plots using this threshold). The severity thresholds were defined in order to maximize the correct classification of extreme severity. The best correct classification defined was 81% of charred plots and 59% of green-scorched plots. To obtain the threshold between unburned and burned forest pixels, we applied the same procedure to a selection of pixels from outside the fire. We also defined a burned threshold to shrub lands and crops to create a continuous map. We applied the thresholds defined in the Ventallo´ fire to the other fires and, together with using information from photographs obtained during and after the wildfires, firefighter reports and personal attendance to the fire events, we defined the final threshold limits by increasing the original threshold computed for Ventallo´ (Figure S1) by 5%. From the information generated for the different fires we estimated the following severity variables in the two levels according to the two objectives: (i) in the first level, among fires (considering the 15 wildfires together), the variable was the percentage of forest pixels burned with extreme severity in the fire; PLOS ONE | www.plosone.org

Fuels, topographic and fire behavior variables In this section we reversed the order of the levels of variables, since many of the variables used in the first level (among fires) were calculated as average values of the variables used for the second level (within fires). We divided the independent variables considered at the two levels in three groups (Table 1). (1) Fuel distribution variables. Fuels are a key factor determining fire severity [44,45]. This group includes variables related to fuels that can directly influence fire severity, such as crown biomass, [17] or indirectly through climatic characteristics that determine fuel quantity and distribution [33]. The variables in the within fires level are: a)

b)

3

Crown biomass (tons/ha), containing branches and leaves without the trunk. We used the crown biomass data of P. halepensis plots from the third National Forest Inventory of Spain [46]. As inventory data was not enough to cover the entire burned area, we tested the relationship between crown biomass and different spectral bands and combinations of bands that had been shown to assess better crown biomass [17] (Table S2). We selected the index showing the highest R2: MID57 (band 5 + band 7); R2 = 0.400, p,0.001, B0 = 42.177 and B1 = 20.061 (exponential regression), after the elimination of the outliers with standardized residual higher than 2 (Figure S2 and Table S3). We applied this relationship to the MID57 values of the studied fires to obtain the crown biomass value of each pixel. Annual Water Availability Index (WAI), obtained from the Digital Climatic Atlas of Catalonia [36,37] and following the January 2014 | Volume 9 | Issue 1 | e85127

Factors Determining Extreme Fire Severity Patterns

Table 1. Independent variables of the three groups considered at the two levels: fuel distribution, topography and fire behavior, analyzed within and among fires.

Group of variables

Within fires

Among fires

Fuel distribution

Crown biomass (tons/ha)

Mean crown biomass (tons/ha)

Water Availability Index (WAI)

Area with crown biomass .15 tons/ha (%)

Drought Code

Climate type Mean Water Availability Index Mean Drought Code

Topography

Slope (u)

Mean slope (u)

Aspect (south/north)

Area with slope higher than 20u (%)

Elevation (m)

Area with southern aspect (%)

Type of slope (upslope/downslope)

Fire type (topographic, convective or wind-driven)

Alignment (full alignment/out of alignment)

Temperature .30uC (yes/no)

Mean elevation (m) Fire behavior

Relative humidity ,30% (yes/no) Wind speed .30 km/h (yes/no) Urban interface within the fire (yes/no) Relative humidity recovered (yes/no) Area with upslope (%) Area with full alignment (%) Units (for continuous variables) or levels (for categorical variables) are shown in brackets. doi:10.1371/journal.pone.0085127.t001

using the available data of all the years before the fire from the meteorological station closest to each fire.

expression   P{PET  100 WAI~ PET

c)

(2) Topographic variables. From a Digital Elevation Model we selected topographic variables that are known to influence fire severity in many previous studies [16,20], such as slope, aspect and elevation [19,29]. Particularly, in 20u slopes head fire spread rates increase four times compared to flat terrain [50]. The variables in the within fires level are:

where P is rainfall (mm yr21) and PET is potential evapotranspiration (mm yr21). Negative values corresponded to dry sites and positive values to wet sites. Drought Code, a component of the Fire Weather Index that was calculated for each pixel following the methods of Van Wagner and Picket [47]. The required daily meteorological data of ten years before the date of ignition was previously corrected using elevation ranges.

a) b) c)

The variables in the among fires level are:

The variables in the among fires level are: a) b)

c)

d)

e) f)

a)

Percentage of forest area, determined from the Land Cover Map of Catalonia [48]. Mean crown biomass, determined as the average value of crown biomass from all forested pixels within the burned area. Percentage of area with crown biomass .15 tons/ha, as crown biomass affects fire severity especially in dense forests with crown biomass over 15 tons/ha [17], determined from crown biomass at the pixel level. Climate type, according to the Thornthwaite moisture index (1948) classification [49]. We considered two main climates in the study area: semi-arid and not semi-arid (including drysubhumid and humid). Mean WAI, the mean WAI of the total number of pixels of the fire. Drought code, developed by the Forest Fire Prevention Service of Catalonia following Van Wagner and Pickett [47]

PLOS ONE | www.plosone.org

Slope, in degrees. Aspect: south (90–270u) and north (270–90u). Elevation, in m.

b) c) d)

Mean slope (u), the mean slope of the total number of pixels of the fire. Percentage of area with slope higher than 20u. Percentage of area with southern aspect (90–270u). Mean elevation (m), the mean elevation of the total number of pixels of the fire.

(3) Fire behavior variables. We included variables related to fire risk and potential fire behavior: type of slope, alignment of factors, fire type, meteorological variables and presence of wildland-urban interface. The basic fire spread factors included were wind and slope, and their coincidence in favor of or against fire led to the concept of alignment of forces [29]. We also considered the basic rule of 30-30-30, which means that relative humidity under 30%, surface wind speed higher than 30 km/h and temperatures higher than 30uC facilitate the increase of fire spread [51].

4

January 2014 | Volume 9 | Issue 1 | e85127

Factors Determining Extreme Fire Severity Patterns

At the among fires level, and given the high number of independent variables in relation to the number of cases, we carried out a multifactorial analysis to reduce the number of independent variables. For that reason, the independent variables were grouped in three groups in relation to fuel distribution, topography and fire behavior. We carried out separate Principal Coordinate Analyses (PCoA) to fuel distribution and fire behavior variables, and a Principal Component Analysis (PCA) to topographic variables. The results obtained (Figures S3, S4, S5) allowed us to select the variables explaining the highest variability in the first axes of the analysis of each of the three groups. We finally selected climate type and mean crown biomass from the fuel distribution group, alignment, urban interface and wind speed .30 km/h from the fire behavior group, and mean slope from the topography group. We carried out a General Linear Model (GLM) to evaluate the effect of these variables, together with fire type, on the proportion of the wildfire surface burned with highest severity. As this variable did not have a normal distribution, we transformed it by the arcsine of the square root to reach normality. Significance was assessed at p,0.05. At the within fires level, as the number of pixels within the burned area of each fire was extremely high, we randomly selected 500 pixels (or all when there were fewer than 500 pixels in any level) of each of the two levels (not extreme severity/extreme severity) of the dependent variable. Previous correlation analysis among the explanatory variables allowed us to eliminate one of each pair of variables that were highly correlated (i.e., Pearson r$0.9). With this procedure we excluded from the analyses Drought Code, elevation and type of slope (correlated with WAI and alignment of factors). Thus, the independent variables considered for these analyses were crown biomass, WAI, alignment, aspect and slope. For each fire separately, a Generalized Linear Model (GLZ) was carried out with binomial distribution and a logit link function. In each analysis, we included the main effects of the variables and the first order interactions. Stepwise model selection was applied starting from the saturated model and removing the least significant term, starting by the interactions until there was no further decrease in the Bayesian Information Criterion (BIC). We considered all models within 2 BIC units as equivalent in terms of fit.

The variables in the within fires level are: a)

b)

Type of slope. We defined two categories depending on slope (upslope or downslope) and wind effect (following wind or low wind effect). We determined the main wind direction of the fire from firefighter reports. Afterwards, the type of slope was defined from the closest topographic wind conditions, that is, from slope and aspect. Thus, the two categories defined were (i) upslope, when wind direction was following upslope, then the aspect of the slope was opposed to wind direction (at 135u of range with the wind direction as a center); and (ii) downslope, when there was leeward slope and the aspect was the same as wind direction. Alignment of factors: combination of type of slope (upslope or downslope) and slope .6u (more than a gentle slope), which was considered the minimum for fire alignment, resulting in two categories: (i) full alignment (upslope with wind following and slope.6u) and (ii) out of alignment (all the other combinations). The variables in the among fires level are:

a)

b)

c)

d)

e)

f)

g) h)

Fire type, with three categories: (i) topographic fire, (ii) convection-dominated fire and (iii) wind-driven fire, according to the classification of the European project ‘‘Fire Paradox’’ [27] and obtained from the firefighter reports and personal attendance during the fires. Temperature .30uC during the fire, with two categories: (i) temperature over 30uC (yes) and (ii) temperature not over 30uC during the fire (no), obtained from the firefighter reports. Relative humidity ,30% during the fire, with two categories: (i) relative humidity over 30% (yes) and (ii) relative humidity not over 30% during the fire (no), obtained from the firefighter reports. Wind speed .30 km/h during the fire, with two categories: (i) wind speed over 30 km/h (yes) and (ii) wind speed not over 30 km/h during the fire (no), obtained from the firefighter reports. Relative humidity recovered (higher than 60% the last night before the wildfire), considering two categories: (i) relative humidity recovered (yes) and (ii) relative humidity not recovered (no) (Official Firefighter reports from the wildfires analyzed). Presence of Wildland-urban interface. From the Firefighter reports we identified the fires within wildland-urban interface, i.e. those fires in which more resources were addressed to protecting property and people than to containing the wildfire, thus affecting their behavior and severity [29]. This variable had two categories: (i) fire in wildland-urban interface (yes) and (ii) fire not in wildland-urban interface (no). Type of slope: % of pixels with upslope and following wind within the burned area. Alignment of factors: % of pixels with full alignment (upslope and slope.6u).

Results Variables affecting fire severity among fires The General Linear Model carried out at the wildfire scale (R2 = 0.81; F = 10.4; p = 0.0013) showed that climate and slope significantly affected fire severity (Table 2). In the case of climate, fires that occurred in non- semi-arid climate showed more extreme fire severity than fires in semi-arid climate (0.7160.04 and 0.2160.07 parts per unit, respectively). Concerning slope, the relationship of this variable with the percent of the area burned with extreme severity was positive, indicating that fires on steeper slopes had more extreme severity than those on gentle slopes.

Variables affecting fire severity within fires Table 3 summarizes the GLZ analyses carried out separately with the fifteen fires considered in this study. Regarding topographic variables, aspect was significant in 73% of fires (71% of topographic, 75% of convective and 75% of wind-driven fires) (Table 3), showing in all but one case more extreme severity in northern than in southern aspects. Slope was significant in 73% of fires, with more extreme severity on steeper than on gentle slopes in all fires except one (Table 3). The interaction between these two topographic variables was significant in 53% of fires,

Data analysis According to the objectives of the study, there were two levels of data analysis: among fires and within fires. The software used was SPSS 17.0 (SPSS Inc., Chicago, IL, USA) and R (version 2.13.2, [52]).

PLOS ONE | www.plosone.org

5

January 2014 | Volume 9 | Issue 1 | e85127

Factors Determining Extreme Fire Severity Patterns

to identify trends about size, frequency and percentage of high severity in fires [53,54]. In our study, the innovative integration of different sources of information (digital cartography, firefighter reports, Landsat images) from historical wildfires across a range of climatic conditions, allowed us to determine patterns and variables affecting extreme fire severity at different levels, among and within fires.

Table 2. Statistical results from the General Linear Model for the effects of climate, wildland-urban interface, crown biomass and slope on extreme severity.

Source of variation

SS

d. f.

MS

F

p

Climate

0.305

1

0.305

11.932

0.006

Wildland-urban interface

0.019

1

0.019

0.734

0.412

Variables affecting fire severity among fires

Crown biomass

0.008

1

0.008

0.305

0.593

Slope

0.140

1

0.140

5.461

0.042

Error

0.256

10

0.026

The analysis among fires shows that climate and slope are the main variables affecting the degree of extreme fire severity (Table 2). The linkage between climate and fire severity is widely known [26,55]. Climate influences fire regime and forest growth through primary productivity, which is related to fuel accumulations that influence fire behavior and fire severity [45,56,57]. Previous studies have shown that the increase in wildfire frequency and fire severity was associated to dry climates, as drought decreases fuel moisture and, consequently, fuel flammability increases [58,59]. But we found the opposite effect, indicating that fires in less arid climates had more extreme severity than those that occurred in more arid climates. In the Mediterranean area, plant growth is primarily constrained by water availability [60,61], suggesting in our study that less arid climates had enough water for forest growth that facilitated crown fires and extreme fire severity. Contrarily, it is accepted that dry conditions reduce vegetation growth, thus leading to a lower accumulation of fuels in terms of quantity and continuity [62]. Specifically, the forest structure predominant in arid conditions is characterized by open forest with very low horizontal continuity of tree crowns , where fire is not likely to spread as an active crown fire, thus reducing extreme fire severity [33]. This was supported by Figure S6, where crown biomass was significantly lower in a semi-arid climate than in a less arid one. Furthermore, the percentage of forest was higher in the less arid climate, which also had more canopy closure (Figure S7). In forest structures characterized by high canopy closure, the risk of active crown fires is higher because there is an increase in vertical and horizontal continuity [3,23,63]. Although some studies have shown that, especially under extreme climatic conditions, climate exerts a dominant control on fire severity and fire behavior, the role of topographic features could also be important [64,65]. Our study shows that slope is significantly higher in fires with more extreme severity. As slope increases, the distance and angle between the flame and fuels is shorter, facilitating the pre-heating of the fuel ahead of the fire front, and thus increasing fire spread [29]. Previous research by different authors has shown that fire severity increased with slope in the direction of fire progression [66–69], while the work of Lentile et al. [70] suggested that on steeper slopes the crowns of large trees were stacked up, thus facilitating fire spread from crown to crown and resulting in more extreme fire severity.

SS, sum of squares; MS, mean square. Significant results at p,0.05 are indicated in bold. doi:10.1371/journal.pone.0085127.t002

showing on steeper slopes more extreme severity in northern than in southern aspects in 88% of these fires, while on gentle slopes the degree of extreme severity was less different or even higher in southern than in northern aspects (Figure 2). Concerning fuels, crown biomass was significant in 93% of fires (86% of topographic and 100% of convective and wind-driven fires). The positive relationship was dominant, indicating more extreme severity at higher crown biomass, especially in topographic and wind-driven fires (Table 3). WAI was significant in 73% of fires (71% of topographic, 100% of convective and 50% of wind-driven), but the direction of the relationship was not homogeneous. The positive relationship predominated in topographic fires (60%), showing that at more WAI there was more extreme severity (Table 3). In convective and wind-driven fires that showed significant WAI, only half of the fires showed a positive relationship. The interaction between aspect and crown biomass was significant in 40% of the fires and it was only relevant in convection-dominated fires (75% of cases), indicating more extreme severity in northern than in southern aspects with high crown biomass, but small differences in extreme severity between aspects with lower crown biomass (Figure 3). The interaction between crown biomass and WAI was significant in 53% of fires (57% of topographic, 50% of convective and 50% of wind-driven) (Table 3). The pattern was similar in all fires except one, showing more extreme severity in higher crown biomass and higher WAI (Figure 4). The interaction between slope and WAI was significant in 47% of fires (28% of topographic, 50% of convective and 75% of wind-driven) (Table 3). But the direction of the relationship was not clear; although 57% of these fires had more extreme severity at low slope and high WAI, this increase was very slight (Figure 5). The remaining interactions were not considered, as the number of significant cases was lower than 33% and the relationship was not clear (Table 3). Finally, alignment was significant in 67% of fires (100% of topographic, 25% of convective and 50% of wind-driven). The negative relationship was predominant in topographic fires (71%), showing more extreme severity when the fire was out of alignment. The relationship was positive in the two wind-driven fires that showed significant alignment (Table 3).

Variables affecting fire severity within fires In Catalonia, where this study has been carried out, there are areas dominated by different fire types, depending on fire spread pattern (topographic, convection, wind-driven) [29]. The variables affecting fire severity among fires were also significant in most of the fires within fires level, but there were more variables determining extreme fire severity at this particular level of analysis. The topographical variables of aspect and slope are key variables in the majority of fires, showing that there is more extreme severity in northern aspects and on higher slopes. Some previous studies have shown that, in the Northern Hemisphere, southern aspects tend to burn with greater intensity and the resultant severity was higher than in other aspects, as southern

Discussion The spatial distribution of fire severity immediately after fire is a key question for validating fire risk maps, fire behavior models and fuel management effectiveness [11]. Previous works in the U.S.A. carried out with extensive databases using historical wildfires at national and regional scales have shown the utility of these studies PLOS ONE | www.plosone.org

6

January 2014 | Volume 9 | Issue 1 | e85127

PLOS ONE | www.plosone.org

7

21.70

0.026

0.000

sig

0.000

0.042

0.027

0.67

0.000

0.031

sig

ALI x WAI B

13.6

4.6

0.03

0.000

17.4

20.08

0.000

15.0

24.50

0.000

141.0

0.24

0.000

15.0

1.30

0.003

9.0

218.00

0.34

0.023

5.1

0.04

0.037

4.3

0.07

0.001

10.8

14.00

0.003

8.8

20.84

0.006

7.5

1.90

20.14

0.000

48.0

0.06

0.000

28.0

20.12

0.023

5.1

25.55

0.000

21.1

20.69

0.000

60.0

3.00

0.008

7

27.40

Castellbisbal Vimbodı´ Nava`s 2005 2006 2007

Wald

0.02

0.000

0.000

SLP x WAI B

38.0

71.0

20.04

0.000

14.0

sig

20.20

0.000

23.0

20.11

0.000

13.7

23.30

0.001

11.7

0.46

0.015

5.8

20.92

0.000

43.0

0.30

0.000

13.6

20.95

Talamanca 2003

Wald

BIO x WAI B

4.1

4.8

20.47

sig

3.70

0.015

5.9

Wald

ASP x BIO B

sig

Wald

0.72

0.014

0.003

sig

ASP x SLP B

5.9

8.7

Wald

20.85

0.000

sig

25.50

63.0

60.0

Wald

2.30

0.000

sig

0.50

0.000

137.0

Wald

B

68.0

1.70

B 29.10

4.9

30.0

Wald

1.02

0.13

0.000

0.019

sig

B

13.8

5.5

Wald

225.00

239.00

B

Alignment B (ALI)

WAI

Crown Biomass (BIO)

Slope (SLP)

Aspect (ASP)

Bot 2000

Castellbisbal Castellbell i 2002 el Vilar 2003

TOPOGRAPHIC FIRES

0.007

7.1

20.07

0.000

19.0

0.47

0.039

4.2

1.80

0.012

6.3

0.52

0.015

5.8

23.10

0.000

41.0

0.08

0.000

20.0

23.70

Margalef 2005

0.000

29.0

20.08

0.000

47.0

1.12

0.000

26.7

23.10

0.002

10.0

0.34

Rocafort 2005

0.09

0.000

35.0

0.01

0.002

9.0

0.13

0.000

25.0

20.22

0.000

35.0

0.25

0.000

50.0

0.68

0.008

6.9

21.80

Cardona 2005

CONVECTION-DOMINATED FIRES

0.005

8.0

0.01

0.000

22.0

0.33

0.008

6.9

20.07

0.019

5.5

20.22

0.000

80.0

0.32

0.003

-

0.61

0.000

14.0

24.40

Castellnou de Bages 2005

Table 3. Statistical results from the Generalized Linear Model for the main effects and interactions on extreme severity in each fire.

0.018

5.0

0.01

0.000

17.0

20.13

0.000

85.0

2.00

0.000

13.0

26.90

0.006

7.4

1.10

0.030

8.5

21.09

Riba-roja d’Ebre 2005

0.81

0.002

9.0

20.86

0.000

20.0

23.00

0.000

182.0

0.37

0.000

18.0

1.50

Cistella 2006

WIND-DRIVEN FIRES

0.004

8.0

0.04

0.000

18.0

0.07

0.001

11.0

20.88

0.000

21.0

2.20

0.001

10.0

1.30

0.039

4.2

20.48

0.044

4.0

0.28

0.000

18.0

21.10

0.001

10.1

3.10

0.010

6.7

0.32

0.042

4.1

222.40

0.008

7.0

20.91

Mont-roig Ventallo´ del Camp 2006 2007

Factors Determining Extreme Fire Severity Patterns

January 2014 | Volume 9 | Issue 1 | e85127

PLOS ONE | www.plosone.org

20.05

0.33

0.09

8 0.000

0.002

B, Wald statistic and significance at p,0.05 are shown. doi:10.1371/journal.pone.0085127.t003

38.0

9.8

sig

20.02

Wald

BIO x SLP B

sig

Wald 0.004

8.2 0.000

13.0

0.000

36.0

20.01

0.044

0.000

sig

ALI x SLP B

4.0

0.10

0.026

4.9

Cardona 2005

21.0

Rocafort 2005

Wald

20.13

0.150

0.005

ALI x BIO B

5.9

7.8

21.19

sig

0.002

10.0

20.20

0.013

6.1

Margalef 2005

CONVECTION-DOMINATED FIRES

Wald

1.00

0.002

0.000

sig

ASP x ALI B

10.0

15.0

20.51

Wald

20.59

0.001

ASP x WAI B

12.0

0.000

Castellbisbal Vimbodı´ Nava`s 2005 2006 2007 17.0

Talamanca 2003

sig

Bot 2000

Castellbisbal Castellbell i 2002 el Vilar 2003

TOPOGRAPHIC FIRES

Wald

Table 3. Cont.

0.002

9.0

0.10

Castellnou de Bages 2005

Riba-roja d’Ebre 2005

0.000

17.0

20.27

0.00

13.0

Cistella 2006

WIND-DRIVEN FIRES

0.010

6.6

21.90

Mont-roig Ventallo´ del Camp 2006 2007

Factors Determining Extreme Fire Severity Patterns

January 2014 | Volume 9 | Issue 1 | e85127

Factors Determining Extreme Fire Severity Patterns

Figure 3. Significant interactions of Aspect (north or south) and Crown Biomass (tons/ha) on extreme fire severity (parts per unit). Letters in the top-right corner indicate the fire: (A) Bot 2000; (B) Castellbisbal 2002; (C) Margalef 2005; (D) Cardona 2005; (E) Castellnou de Bages; (F) Mont-roig del Camp 2007. Letters in the bottom-left corner indicate the fire type: (T) topographic fires; (C) convective fires; (W) wind-driven fires. doi:10.1371/journal.pone.0085127.g003

Figure 2. Significant interactions of Aspect (north or south) and Slope (degrees) on extreme fire severity (parts per unit). Letters in the top-right corner indicate the fire: (A) Bot 2000; (B) Castellbell i el Vilar 2003; (C) Castellbisbal 2005; (D) Vimbodı´ 2006; (E) Nava`s 2007; (F) Castellnou de Bages 2005; (G) Cistella 2006; (H) Montroig del Camp 2007. Letters in the bottom-left corner indicate the fire type: (T) topographic fires; (C) convective fires; (W) wind-driven fires. doi:10.1371/journal.pone.0085127.g002

Alignment of factors is a complex variable that cannot usually be estimated, as there is no information about variables such as wind direction at landscape scale. In our study, alignment of factors was significant in nearly 67% of wildfires, but the pattern was not homogeneous and higher alignment of factors implied higher extreme fire severity in only 50% of the cases. Although we expected that fire severity would increase with the combination of high upslope and wind direction, the opposite trend was observed for topographic fires and no trend was observed for convective and wind-driven fires. This is likely due to other factors that could modify the pattern of fire severity, such as forest structure, fuel moisture or the type of spread (head, flank or back fire), thus not all the areas aligned burned with the same severity. In topographic fires, we found the opposite trend because wind is not the most important factor explaining fire severity and behavior, as these fires are caused by complex relationships among fuel heating, slope and topographical winds [29]. Topographic fires showed a dominant pattern of more extreme severity at higher values of WAI (Table 3). This confirms the pattern mentioned among fires, where lower water limitations could lead to higher crown biomass and continuity that increase the proportion of extreme severity [12,69]. In wind-driven and convective fires, this evidence was not clear probably because other variables were interacting at the same time. Wind-driven fires are characterized by high wind speed that could minimize the effect of WAI in determining extreme fire severity. In the same way, convective fires generate their own fire environment, thus reducing the effect of other local variables on spread [29]. Although WAI as a main variable was not significant in convective fires, in this type of fire there was more extreme severity in

aspects received more solar radiation and fuel moisture was lower [65,71]. But in southern aspects growth limitations are also common and for this reason the effect of topographical variables is not always coincident [22,72]. The results of our study indicate that extreme severity was more frequent in northern aspects (Table 3), where solar radiation was lower and fuels had less water limitation to growth, which could promote higher levels of tree density and vertical growth, as was shown in the Al Omary study [73]. This later pattern explains why northern aspects have higher crown biomass, promoting high fuel continuity and more extreme severity [19,69,73]. Slope showed the same tendency as in the previous among fires analysis, with more extreme severity on steeper slopes. The interaction between slope and aspect was also significant especially in topographic fires for a similar reason: the degree of extreme severity was highest on steeper slopes and northern aspects (Figure 2), as in this combination of variables fire spread and fuel continuity could be the highest. These findings were supported by the fuel results of the GLZ models, since there was more extreme severity at high values of crown biomass, especially in topographic and wind-driven fires (Table 3). Higher values of crown biomass affect fire behavior by increasing the area affected by crown fires and producing more extreme severity [32,74].

PLOS ONE | www.plosone.org

9

January 2014 | Volume 9 | Issue 1 | e85127

Factors Determining Extreme Fire Severity Patterns

Figure 4. Significant interactions of Crown Biomass (tons/ha) and WAI on extreme fire severity (parts per unit). Letters in the top-right corner indicate the fire: (A) Castellbell i el Vilar 2003; (B) Talamanca 2003; (C) Vimbodı´ 2006; (D) Nava`s 2007; (E) Margalef 2005; (F) Rocafort 2005; (G) Riba-roja d’Ebre 2005; (H) Ventallo´ 2006. Letters in the top-left corner indicate the fire type: (T) topographic fires; (C) convective fires; (W) wind-driven fires. doi:10.1371/journal.pone.0085127.g004

Figure 5. Significant interactions of Slope (6) and WAI on extreme fire severity (parts per unit). Letters in the top-right corner indicate the fire: (A) Castellbell i el Vilar 2003; (B) Castellbisbal 2005; (C) Cardona 2005; (D) Castellnou de Bages 2005; (E) Riba-roja d’Ebre 2005; (F) Ventallo´ 2006; (G) Mont-roig del Camp 2007. Letters in the top-left corner indicate the fire type: (T) topographic fires; (C) convective fires; (W) wind-driven fires. doi:10.1371/journal.pone.0085127.g005

northern than in southern aspects with high crown biomass that can be related to water content (Figure 3). Convective fires are dominated by the accumulation of highly available fuel that is higher at northern aspects [25]. Moreover, even though there was more extreme severity at high crown biomass, this increase was much faster at higher values of WAI, especially in topographic fires (Figure 4). In less arid environments (where WAI is higher), flammability is lower and the quantity of heat necessary to start the ignition is higher [59]. Nevertheless, when fuels are available for burning under extreme meteorological conditions, they can burn as high intensity crown fires, thus involving more extreme severity [26,75]. This pattern has recently been described in the Mediterranean area, where fires of high intensity and severity occur in humid forests that are not very prone to burn, such as some montane (sub-Mediterranean) areas of Pinus nigra and Pinus sylvestris [76,77] and high mountain forests of Pinus uncinata in the Pyrenees ([78], Albert Alvarez, personal observation). PLOS ONE | www.plosone.org

Limitations and future implications The analysis of historical wildfires is used to identify the main type of spread pattern in each area of a country and, thus, to know where it is more probable to have one of the three fire types. The probability to have a type of fire can be mapped and it is useful for firefighters to know the strategies, opportunities and critical points that they should use to control the fire before it starts. The use of the fire type concept is also useful to drive fuel management strategies through the definition of the Strategic Management Points (SMP), which are the key points where it is necessary to create infrastructures to limit the extent of large wildfires [29]. The use of remotely sensed data and their transferability to the dNBR index allowed us to determine the variables and patterns affecting fire severity using historical wildfires that could not be studied by other means. Our results suggest that this methodology could be 10

January 2014 | Volume 9 | Issue 1 | e85127

Factors Determining Extreme Fire Severity Patterns

Figure S5 Principal Component Analysis (PCA) of topographic variables. (TIFF)

applied to other fires in the Mediterranean area in a systematic way, but information from the ‘‘real’’ fire severity from some wildfires is needed, using direct data from firefighters or from postfire plots in order to improve the accuracy in the definition of the severity thresholds and the validation of final fire severity maps. We also believe that the methodology used in this work could be improved by including additional information from forest structures. The use of Landsat time-series metrics, LIDAR or radar remote-sensing instruments (e.g., Pol-SAR) has recently been suggested as an effective way to provide information from aboveground biomass and forest structures [79,80]. The combination of these data sources could improve the methodology in detecting specific areas such as the flanks of the fire runs that were probably not identified correctly. Global climate change is increasing the number, severity and recurrence of forest fires, as well as the surface burned and the length of the fire risk season in the Mediterranean Basin [81]. Water scarcity and the expected increase of extreme droughts caused by climate warming will lead to an increase in highintensity crown fires in areas that have not traditionally been subject to this type of fire ([78], Albert Alvarez, pers. obs.). Consequently, future research and management strategies to reduce future wildfires with crown fires and extreme fire severity should be focused not just on the most commonly burned areas but also on the less usually burned, since these new zones with high crown biomass accumulated in non-common drought areas will be available to burn as extreme severity wildfires.

Figure S6

Relation between climate and biomass (tons/

ha). (TIFF) Figure S7

Relation between climate and forested area

(%). (TIF) Table S1 Information about the fires analyzed. Fire name, date of ignition, path/row, pre-fire data, post-fire data, sensor of the images, fire size, range of elevation, range of slope and fire type. (PDF) Table S2 Bands related to Crown Biomass, regression type, B0(constant), B1(regression coefficient) and R square. (PDF) Table S3 Parameter estimates of the regression be-

tween Crown Biomass and MID57 index. (PDF)

Acknowledgments We wish to thank members of the Centre for Ecological Research and Forestry Applications (CREAF) for their technical help: X. Arnan, J. Iban˜ez, R. Molowny, G. More´, D. Pascual, E. Pla and J. Vayreda. We are also grateful to M. Ninyerola, the Meteorological Service of Catalonia (SMC) and the State Meteorological Agency (AEMET), which facilitated meteorological data. We are also grateful for the contribution of F. X. Castro (Forest Fires Prevention Services, Generalitat de Catalunya) in the Drought Code assessment among fires, and J. Minett for English language consultation. The Generalitat de Catalunya fire-fighters corps shared information from wildfires and provided useful comments based on experience in fire fighting, in particular M. Castellnou and M. Miralles. R.W. Gray and an anonymous reviewer provided constructive comments on the manuscript.

Supporting Information Figure S1 dNBR thresholds limits for forest (A), shrublands and fruit tree crops (B) and other crops (C). (TIF) Figure S2 Exponential regression between Crown Biomass (tons/ha) and MID57 index. (TIFF)

Principal Coordinate Analysis (PCoA) of fuel distribution variables. (TIFF)

Figure S3

Author Contributions

Figure S4 Principal Coordinate Analysis (PCoA) of fire

Conceived and designed the experiments: JLD AA JR. Performed the experiments: JLD AA. Analyzed the data: JLD AA JR. Contributed reagents/materials/analysis tools: JLD AA JR. Wrote the paper: JLD.

behavior variables. (TIFF)

References 9. Miller JD, Thode AE (2007) Quantifying burn severity in a heterogeneous landscape with a relative version of the delta Normalized Burn Ratio (dNBR). Remote Sens Environ 109: 66–80. 10. Key CH, Benson NC (2006) Landscape assessment (LA): sampling and analysis methods. In FIREMON: Fire Effects Monitoring and Inventory System. General Technical Report RMRS-GTR-164-CD. USDA Forest Service, Rocky Mountain Research Station, Fort Collins, Co. 11. Stow D, Petersen A, Rogan J, Franklin J (2007) Mapping burn severity of mediterranean-type vegetation using satellite multispectral data. GIsci Remote Sens 44: 1–23. 12. Alvarez A, Gracia M, Castellnou M, Retana J (2013) Variables That Influence Changes in Fire Severity and Their Relationship with Changes Between Surface and Crown Fires in a Wind-Driven Wildfire. For Sci 59: 139–150. 13. Lentile LB, Holden ZA, Smith AMS, Falkowski MJ, Hudak AT, et al. (2006) Remote sensing techniques to assess active fire characteristics and post-fire effects. Int J Wildland Fire 15: 319–345. 14. Schulz JJ, Cayuela L, Echeverria C, Salas J, Benayas JMR (2010) Monitoring land cover change of the dryland forest landscape of Central Chile (1975–2008). Appl Geogr 30: 436–447. 15. Alrababah MA, Alhamad MN (2006) Land use/cover classification of arid and semi-arid Mediterranean landscapes using Landsat ETM . Int J Remote Sens 27: 2703–2718.

1. Pausas JG, Vallejo VR (1999) The role of fire in European Mediterranean Ecosystems. In: Chuvieco E, editor. Remote sensing of large wildfires in the European Mediterranean basin. Berlin: Springer-Verlag. pp. 3–16. 2. Scott JH, Reinhardt ED (2001) Assessing Crown Fire Potential by Linking Models of Surface and Crown Fire Behavior. Research Paper RMRS-29. USDA Forest Service, Rocky Mountain Research Station, Fort Collins, Co. 3. Van Wagner CE (1977) Conditions for the start and spread of crown fire. Can J For Res 7: 23–24. 4. Covington WW (2000) Helping western forests heal: the prognosis is poor for US forest ecosystems. Nature 408: 135–136. 5. Fule´ PZ, Covington WW, Smith HB, Springer JD, Heilein TA, et al. (2002) Comparing ecological restoration alternatives: Grand Canyon, Arizona. For Ecol Manage 170: 19–41. 6. Werth PA, Potter BE, Clements CB, Finney MA, Goodrick SL, et al. (2011) Synthesis of Knowledge of Extreme Fire Behavior: Volume I for Fire Managers. General Technical Report PNW-854. USDA Forest Service, Pacific Northwest Research Station, Portland, Or. 7. Stocks BJ, Alexander ME, Lanoville A (2004) Overview of the International Crown Fire Modelling Experiment (ICFME). Can J For Res 34: 1543–1547. 8. Taylor SW, Wotton BM, Alexander ME, Dalrymple GN (2004) Variation in wind and crown fire behaviour in a northern jack pine black spruce forest. Can J For Res 34: 1561–1576.

PLOS ONE | www.plosone.org

11

January 2014 | Volume 9 | Issue 1 | e85127

Factors Determining Extreme Fire Severity Patterns

43. Pons X (2000) MiraMon. Sistema d’Informacio´ Geogra`fica i software de Teledeteccio´. Centre for Ecological Research and Forestry Applications (CREAF). Cerdanyola del Valle`s, Barcelona. 44. Martinson EJ, Omi PN (2008) Assessing mitigation of wildfire severity by fuel treatments - an example from the Coastal Plain of Mississippi. Int J Wildland Fire 17: 415–420. 45. Graham RT, McCaffey S, Jain TB (2004) Science basis for changing forest structure to modify wildfire behavior and severity. General Technical Report RMRS-GTR- 120. USDA Forest Service, Rocky Mountain Research Station, Fort Collins, Co. 46. Villanueva JA (2005) Tercer Inventario Forestal Nacional (1997–2007). Madrid: Ministerio de Medio Ambiente, ICONA. 47. Van Wagner C and Pickett T (1985) Equations and FORTRAN program for the Canadian Forest Fire Weather Index System. Canadian Forestry Service, Otawa, Ont. pp.1–46. 48. MCSC (2005) Mapa de Cobertes del So`l de Catalunya. Centre for Ecological Research and Forestry Applications (CREAF). Cerdanyola del Valle` s, Barcelona. 49. Thornthwaite CW (1948) An Approach toward a Rational Classification of Climate. Geographical Review. American Geographical Society 38: 55–94. 50. McArthur AG (1968) Fire Behaviour in Eucalypt Forests. Ninth Commonwealth Forestry Conference. Leaflet No. 107. 51. Ve´lez R (2000) Los incendios forestales en la Cuenca Mediterra´nea. In: Ve´lez R, editor. La defensa contra incendios forestales. Fundamentos y experiencias. Madrid: MacGraw-Hill. pp. 3.1–3.15. 52. R Development Core Team (2011) R: a language and environment for statistical computing. Vienna, Austria. 53. Schwind B, Brewer K, Quayle B, Eidenshink JC (2010) Establishing a nationwide baseline of historical burn-severity data to support monitoring of trends in wildfire effects and national fire policies. General Technical Report PNW-GTR-802. USDA Forest Service, Pacific Northwest Research Station, Portland, Or. 54. Miller JD, Skinner CN, Safford HD, Knapp EE, Ramirez CM (2012) Trends and causes of severity, size, and number of fires in northwestern California, USA. Ecol Appl 22: 184–203. 55. Westerling AL, Hidalgo HG, Cayan DR, Swetnam TW (2006) Warming and earlier spring increase western U.S. forest wildfire activity. Science 313: 940– 943. 56. Va´zquez A, Garcı´a JM, Ortega M, Sa´nchez O (2006) Recent fire regime in peninsular Spain in relation to forest potential productivity and population density. Int J Wildland Fire 15: 397–405. 57. Ireland KB, Stan AB, Fule´ PZ (2012) Bottom-up control of a northern Arizona ponderosa pine forest fire regime in a fragmented landscape. Landsc Ecol 27: 983–997. 58. Aldersley A, Murray SJ, Cornell SE (2011) Global and regional analysis of climate and human drivers of wildfire. Sci Total Environ 409: 3472–3481. 59. Dimitrakopoulos AP, Papaioannou KK (2001) Flammability Assessment of Mediterranean Forest Fuels. Fire Technol 37: 143–152. 60. Hetherington AM, Woodward FI (2003) The role of stomata in sensing and driving environmental change. Nature 424: 901–908. 61. Raz-Yaseef N, Rotenberg E, Yakir D (2010) Effects of spatial variations in soil evaporation caused by tree shading on water flux partitioning in a semi-arid pine forest. Agr Forest Meteorol 150: 454–462. 62. Vicente-Serrano SM, Lasanta T, Gracia C (2010) Aridification determines changes in forest growth in Pinus halepensis forests under semiarid Mediterranean climate conditions. Agr Forest Meteorol 150: 614–628. 63. Cruz MG, Alexander ME, Wakimoto RH (2004) Modeling the Likelihood of Crown Fire Occurrence in Conifer Forest Stands. For Sci 50: 640–658. 64. Turner MG, Romme WH (1994) Landscape dynamics in crown fire ecosystems. Landsc Ecol 9: 59–77. 65. Alexander JD, Seavy NE, Ralph CJ, Hogoboom B (2006) Vegetation and topographical correlates of fire severity from two fires in the Klamath-Siskiyou region of Oregon and California. Int J Wildland Fire 15: 237–245. 66. Lee B, Kim SY, Chung J, Park PS (2008) Estimation of fire severity by use of Landsat TM images and its relevance to vegetation and topography in the 2000 Samcheok forest fire. J For Res 13: 197–204. 67. Rothermel RC (1972) A mathematical model for predicting fire spread in Wildland fuels. General Technical Report INT-115. USDA Forest Service, Intermountain Forest and Range Experiment Station, Ogden, Ut. 68. Chung JS, Lee BD, Lee SY (2002) Development of a forestland slope interpretation module for predicting landslide hazards. J Korean For Soc 91: 34–41. 69. Holden ZA, Morgan P, Evans JS (2009) A predictive model of burn severity based on 20-year satellite-inferred burn severity data in a large southwestern US wilderness area. For Ecol Manage 258: 2399–2406. 70. Lentile LB, Smith FW, Shepperd WD (2006) Influence of topography and forest structure on patterns of mixed severity fire in ponderosa pine forests of the South Dakota Black Hills, USA. Int J Wildland Fire 15: 557–566. 71. Weatherspoon CP, Skinner CN (1995) An Assessment of Factors Associated with Damage to Tree Crowns from the 1987 Wildfires in Northern California. For Sci 41: 430–451. 72. Gong X, Brueck H, Giese KM, Zhang L, Sattelmacher B, et al. (2008) Slope aspect has effects on productivity and species composition of hilly grassland in the Xilin River Basin, Inner Mongolia, China. J Arid Environ 72: 483–493.

16. Oliveras I, Gracia M, More´ G, Retana J (2009) Factors influencing the pattern of fire severities in a large wildfire under extreme meteorological conditions in the Mediterranean basin. Int J Wildland Fire 18: 755–764. 17. Garcı´a-martı´n A, Pe´rez-cabello F, Ferna´ndez JdlR, Lloverı´a RM (2008) Estimation of Crown Biomass of Pinus spp. From Landsat TM and Its Effect on Burn Severity in a Spanish Fire Scar. IEEE J Sel Top Appl Earth Observ Remote Sens 1: 254–265. 18. Soverel NO, Coops NC, Perrakis DDB, Daniels LD, Gergel SE (2011) The transferability of a dNBR-derived model to predict burn severity across 10 wildland fires in western Canada. Int J Wildland Fire 20: 518–531. 19. Broncano MJ, Retana J (2004) Topography and forest composition affecting the variability in fire severity and post-fire regeneration occuring after a large fire in the Mediterranean basin. Int J Wildland Fire 13: 209–216. 20. Lee S-W, Lee M-B, Lee Y-G, Won M-S, Kim J-J, et al. (2009) Relationship between landscape structure and burn severity at the landscape and class levels in Samchuck, South Korea. For Ecol Manage 258: 1594–1604. 21. Turner MG, Romme WH, Gardner RH (1999) Prefire Heterogeneity, Fire Severity, and Early Postfire Plant Reestablishment in Subalpine Forests of Yellowstone National Park, Wyoming. Int J Wildland Fire 9: 21–36. 22. Chafer CJ, Noonan M, Macnaught E (2004) The post-fire measurement of fire severity and intensity in the Christmas 2001 Sydney wildfires. Int J Wildland Fire 13: 227–240. 23. Alvarez A, Gracia M, Retana J (2012) Fuel types and crown fire potential in Pinus halepensis forests. Eur J For Res 131: 463–474. 24. Kulakowski D, Veblen TT (2007) Effect of prior disturbances on the extent and severity of wildfire in Colorado subalpine forests. Ecology 88: 759–769. 25. Bessie WC, Johnson EA (1995) The Relative Importance of Fuels and Weather on Fire Behavior in Subalpine Forests. Ecology 76: 747–762. 26. Schoennagel T, Veblen TT, Romme WH (2004) The Interaction of fire, fuels, and climate across Rocky Mountain Forests. BioScience 54: 661–676. 27. Silva JS, Rego F, Fernandes P, Rigolot E, editors(2010) Towards Integrated Fire Management - Outcomes of the European Project Fire Paradox. European Forest Institute Research Report 23. Available: http://www.efi.int/files/ attachments/publications/efi_rr23.pdf. Accessed 30 Oct 2013. 28. Castellnou M, Page´s J, Miralles M, Pique´ M (2009) Tipificacio´n de los incendios forestales de Catalun˜a. Elaboracio´n del mapa de incendios de disen˜o como herramienta para la gestio´n forestal. Proceedings ot the 5th Congreso Forestal Espan˜ ol A´ vila, Spain. Available: http://www.incendioforestal.es/index. php?option = com_mtree&task = att_download&link_id = 264&cf_id = 24. Accessed 30 Oct 2013. 29. Costa P, Castellnou M, Larran˜aga A, Miralles M, Kraus D (2011) Prevention of Large Wildfires using the Fire Types Concept. Unitat Te`cnica del GRAF, Cerdanyola del Valle` s, Barcelona. Available http://www.efi.int/files/ attachments/publications/handbook-prevention-large-fires_en.pdf. Accessed 30 Oct 2013. 30. Pausas JG (2004) Changes in fire and climate in the eastern Iberian Peninsula (Mediterranean Basin). Clim Change 63: 337–350. 31. Castellnou M, Larran˜aga A, Miralles M, Vilalta O, Molina D (2010) Improving Wildfire Scenarios: Learning from Experience. In: Silva JS, Rego F, Fernandes P, Rigolot E, editors. Towards Integrated Fire Management - Outcomes of the European Project Fire Paradox. European Forest Institute Research Report 23: 121–133. 32. Mitsopoulos ID, Dimitrakopoulos AP (2007) Canopy fuel characteristics and potential crown fire behavior in Aleppo pine (Pinus halepensis Mill.) forests. Ann For Sci 64: 287–299. 33. Alvarez A, Gracia M, Vayreda J, Retana J (2012) Patterns of fuel types and crown fire potential in Pinus halepensis forests in the Western Mediterranean Basin. For Ecol Manage 270: 282–290. 34. Capel JJ (2000) El Clima de la Penı´nsula Ibe´rica. Barcelona: Ariel. 35. Ninyerola M, Pons X, Roure JM (2005) Atlas Clima´tico de la Penı´nsula Ibe´rica. Metodologı´a y aplicaciones en bioclimatologı´a y geobota´nica. Universidad Auto´noma de Barcelona, Cerdanyola del Valle´s. 36. Pons X (1996) Estimacio´n de la Radiacio´n Solar a partir de modelos digitales de elevaciones. Propuesta metodolo´gica. In: Juaritsi J, Moro I, editors. VII Coloquio de Geografı´a Cuantitativa, Sistemas de Informacio´n Geogra´fica y Teledeteccio´n. Vitoria-Gasteiz. 37. Ninyerola M, Pons X, Roure JM (2000) A methodological approach of climatological modelling of air temperature and precipitation through GIS techniques. Int J Climatol 20: 1823–1841. 38. Pala` V, Pons X (1995) Incorporation of relief into geometric corrections based on polynomials. Photogramm Eng Remote Sensing 61: 935–944. 39. Pons X, Sole´-Sugranes L (1994) A Simple Radiometric Correction Model to Improve Automatic Mapping of Vegetation from Multispectral Satellite Data. Remote Sens Environ 25: 191–204. 40. Coppin P, Jonckheere I, Nackaerts K, Muys B, Lambin E (2004) Digital change detection techniques in ecosystem monitoring: a review. Int J Remote Sens 25: 1565–1596. 41. Escuin S, Navarro R, Ferna´ndez P (2008) Fire severity assessment by using NBR (Normalized Burn Ratio) and NDVI (Normalized Difference Vegetation Index) derived from LANDSAT TM/ETM images. Int J Remote Sens 29: 1053–1073. 42. Keeley JE (2009) Fire intensity, fire severity and burn severity: a brief review and suggested usage. Int J Wildland Fire 18: 116–126.

PLOS ONE | www.plosone.org

12

January 2014 | Volume 9 | Issue 1 | e85127

Factors Determining Extreme Fire Severity Patterns

77. Rodrigo A, Retana J, Pico´ FX (2004) Direct regeneration is not the only response of Mediterranean forests to intense fires. Ecology 85: 716–729. 78. Pausas JG, Llovet J, Rodrigo A, Vallejo R (2008) Are wildfires a disaster in the Mediterranean basin? A review. Int J Wildland Fire 17: 713–723. 79. Pflugmacher D, Cohen WB, Kennedy RE (2012) Using Landsat-derived disturbance history (1972–2010) to predict current forest structure. Remote Sens Environ 122: 146–165. 80. Kane VR, Lutz JA, Roberts SL, Smith DF, McGaughey RJ, et al. (2013) Landscape-scale effects of fire severity on mixed-conifer and red fir forest structure in Yosemite National Park. For Ecol Manage 287: 17–31. 81. Moriondo M, Good P, Durao R, Bindi M, Giannakopoulos C, et al. (2006) Potential impact of climate change on fire risk in the Mediterranean area. Climate Res 31: 85–95.

73. Al Omary A (2011) Effects of aspect and slope position on growth and nutritional status of planted Aleppo pine (Pinus halepensis Mill.) in a degraded land semi-arid areas of Jordan. New Forest 42: 285–300. 74. Cruz MG, Alexander ME, Wakimoto RH (2003) Assessing canopy fuel stratum characteristics in crown fire prone fuel types of western North America. Int J Wildland Fire 12: 39–50. 75. Bradstock RA, Hammill KA, Collins L, Price O (2010) Effects of weather, fuel and terrain on fire severity in topographically diverse landscapes of south-eastern Australia. Landsc Ecol 25: 607–619. 76. Retana J, Espelta JM, Habrouk A, Ordo´n˜ez JL, Sola`-Morales F (2002) Regeneration patterns of three Mediterranean pines and forest changes after a large wildfire in northeastern Spain. Ecoscience 9: 89–97.

PLOS ONE | www.plosone.org

13

January 2014 | Volume 9 | Issue 1 | e85127